From 890fc3dd29681d03ff07c01b548c68d2f956f65c Mon Sep 17 00:00:00 2001 From: brianmmaina Date: Mon, 17 Aug 2026 12:48:23 -0700 Subject: [PATCH] docs: fix a claim the clock correction superseded The Reading section still said the laws fail by up to four orders of magnitude on diffusion. That was the raw A4 clock number and the correction two sections above already showed most of it is a unit mismatch. It survived the earlier sweep because the phrase wraps across a line break, so grep for the whole string missed it. Also sharpened what that section claims. It led with the error being monotonic in dp/p_c, which oversells: the rank correlation is significant for diffusion but not for spread, at n=5 the smallest attainable p is the one we got, and the ordering inside the small tick group is not real since GOOG and AAPL swap in every block. The robust claim is the two group separation, which has 65 non overlapping block measurements behind it. Says that instead. --- docs/FARMER_2005.md | 20 ++++++++++++++++---- 1 file changed, 16 insertions(+), 4 deletions(-) diff --git a/docs/FARMER_2005.md b/docs/FARMER_2005.md index d21966f..deddfa2 100644 --- a/docs/FARMER_2005.md +++ b/docs/FARMER_2005.md @@ -251,12 +251,24 @@ so it is not evidence. **The ratio column and the rank test are the evidence.** The paper's claim, as conditioned by the paper itself, is **not refuted here** — it is confirmed in the weak sense available from five stocks: -1. Across the full sample the laws fail, badly, by up to four orders of - magnitude on diffusion. -2. The failure is **monotonic in `dp/p_c`**, the model's own scope parameter - (p = 0.017 for diffusion). +1. Across the full sample the laws fail badly — spread ratios spanning 3.7 to + 50, and diffusion ratios spanning 66× on a matched clock. (The raw A4 clock + makes diffusion look like 1,777×, but most of that is the unit mismatch + described above, not model failure.) +2. The stocks separate into **two non-overlapping groups** by `dp/p_c`, the + model's own scope parameter, and the separation survives being re-measured + in all 13 intraday blocks. 3. Where the scope condition holds, the spread ratio is roughly constant. +On the strength of the evidence: the **two-group separation** in point 2 is the +robust claim, with 65 non-overlapping block measurements behind it. The rank +correlation is weaker than it looks — significant for diffusion (ρ = 1.000, +p = 0.0167) but *not* for spread (ρ = 0.900, p = 0.0833), and at n = 5 the +smallest attainable p-value is 0.0167 regardless. The ordering *within* the +small-tick group is not real: GOOG and AAPL swap, persistently and in every +block. So the defensible statement is that tick-constrained stocks behave +categorically differently, not that the error is monotonic in `dp/p_c`. + Ignoring the tick parameter turns a scope condition into an apparent refutation. That is the substantive lesson, and it is the same mistake in a new costume as the one at the top of this document: **testing a model outside the